improving cache effectiveness based on cooperative cache...
TRANSCRIPT
Improving Cache Effectiveness Based on CooperativeCache Management in MANETs
Ali Larbi1 • Louiza Bouallouche-Medjkoune1 • Djamil Aissani1
� Springer Science+Business Media, LLC 2017
Abstract In wireless mobile Ad Hoc networks, cooperative cache management is con-
sidered as an efficient technique to increase data availability and improve access latency.
This technique is based on coordination and sharing of cached data between nodes
belonging to the same area. In this paper, we studied the cooperative cache management
strategies. This has enabled us to propose a collaborative cache management scheme for
mobile Ad Hoc networks, based on service cache providers (SCP), called cooperative
caching based on service providers (CCSP). The proposed scheme enabled the election of
some SCPs mobile nodes, which receive cache’s summaries of neighboring nodes. Thus,
nodes belonging to the same zone can locate easily cached documents of that area. The
election mechanism used in this approach is executed periodically to ensure load bal-
ancing. We further provided an evaluation of the proposed solution, in terms of request hit
rate, byte hit rate and time gains. Compared with other caching management schemes, the
simulation results show that the proposed CCSP scheme improves significantly the cache
effectiveness and the network performances. This is achieved by improving data avail-
ability and reducing both overall network load and latencies perceived by end users.
Keywords Mobile Ad Hoc networks � Data service management � Cooperativecaching � Service cache provider � Cache management � Simulation
& Ali [email protected]
Louiza [email protected]
Djamil [email protected]
1 Research Unit LaMOS, ‘‘Modeling and Optimization of Systems’’, Faculty of Exact Sciences,Bejaia University, Targa Ouzemmour, 06000 Bejaıa, Algeria
123
Wireless Pers CommunDOI 10.1007/s11277-017-4984-7
1 Introduction
The last decade has seen a rapid growth in computer and wireless communication tech-
nologies, which stimulated the development of mobile communication systems. Conse-
quently, two categories of mobile networks have been developed: (i) infrastructure based
and (ii) Ad Hoc networks [1]. The infrastructure based network uses fixed network access
points (APs or base stations) with which mobile terminals interact for communicating. APs
are responsible for receiving the signal from end devices and retransmitting them to the
destinations. Its role is also to interconnect the wireless LAN to external networks such as
internet. However, these networks are frequently unsuitable for many applications and
environments (hostile environment, need for a rapid deployment,...), thus, the deployment
of fixed infrastructure is either too expensive, or impossible. Due to their flexibility and
ease of deployment, distributed solutions, known as mobile Ad Hoc NETworks (MAN-
ETs), are potentially more adapted for applications such as battlefield, disaster recovery
environments or outdoor assemblies. MANET must be seen as complimentary of infras-
tructure-based network rather than competitive [2, 3]. Indeed, in infrastructure based, the
network coverage can be extended by Ad Hoc networks.
The posed constraints in the particular context of the mobile Ad Hoc networks are
different, and especially more complex than those encountered in the wired architecture
[4, 5]. These networks may suffer from scarce bandwidth, frequent network disconnections
and limited local resources like memory’s capacity. However, with the spectacular pro-
gress of microelectronics, the storage capacities of memories and the data transfer rates
have significantly improved. Still, this asset remains insufficient in comparison with the
tendency to increase the number of end users and exchanged information volume between
them. This new deal will reduce increasingly the level of quality-of-service required by
mobile end users, particularly their perceived latencies. The use of cooperative caching is
proved to be efficient solution to improve latencies perceived by users and to increase
overall storage capacity of caches [1, 6–10]. The cache refers to the storage mechanism of
usually or recently required information, for a future re-use [5, 9]. The basic idea is due to
the fact of storing the data items in the local cache of the mobile nodes; the mobile Ad Hoc
network members can effectively reach required information without request servers. This
technique has been widely studied mainly to minimize average latency perceived by the
end users, to save bandwidth and to reduce energy consumption [6, 11]. However, in
mobile Ad Hoc networks, nodes belonging to the same geographic area may have similar
tasks and share common interest, thus cache replacement strategies that perform
replacement individually cannot be suitable [12, 13]. Cooperative caching based on the
sharing and coordination of cached data among multiple clients, allows the cache systems
to be optimized, and the use of all network resources can be more effective. There are a
very great number of cache management mechanisms in literature and most of them were
proposed in the field of Web. Cache management techniques designed for wired networks
may not be applicable to wireless Ad Hoc networks. In this paper, a new approach was
established for a cooperative cache management in mobile Ad Hoc networks based on
service cache providers (SCP). This new approach allows the election of some mobile
nodes to be SCPs that receive summaries caches of neighboring nodes. Consequently,
nodes belonging to the same area can reach all requested documents, which are locally
cached by nodes of this zone. We evaluate the performance of our CCSP scheme by
comparing it with CC scheme [14] (it is the ‘‘closer’’ competitor) and SimpleCache
scheme (SC) [15]. We perform an experimental evaluation of these schemes and
A. Larbi et al.
123
simulation results attests that the proposed CCSP scheme can significantly improve data
accessibility and bandwidth saving.
The rest of this paper is organized as follows. Section 2 reviews some works related to
cache replacement policies in MANETs and cooperative caching schemes. Section 3
exposes first the system model and the presentation of used local cache replacement policy.
Then describes the proposed cooperative cache scheme. In Sect. 4, simulation model and
summary of used notations are presented. Section 5 is devoted to the presentation and to
the discussion of obtained simulation results. The concluding remarks are given in Sect. 6.
2 Existing Works
2.1 Cache Replacement Policies in MANETs
The cache replacement policies in MANETs can be either uncoordinated or coordinated, in
case of uncoordinated cache replacement policy, the selection process of documents to
delete from the cache is made based only on document’s characteristics and local cache
state. However, in case of coordinated cache replacement policy, In addition to previous
parameters, the selection process take into account all cache states of nodes belonging to
the same managed zone. Thus, all local caches of this zone will be managed as a single
large cache.
There is a very great number of cache replacement policies in literature, most of them
were proposed in the field of Web. However, those policies do not take into account the
characteristics of MANETs like mobility. Among the cache replacement policies, which
were proposed within the framework of these networks, we can quote the TDS policy (time
and distance sensitive) reported in [16]. This policy takes into account a distance factor and
a temporal factor for the replacement. TDS calculates a value v(i) for each document i,
vðiÞ ¼ di � si, where di is the distance measured by the number of hops towards the access
points or mobile terminals, having the requested data. si ¼ 1=ðtcur � tepdateiÞ, where tcur and
tepdateiare respectively the current time and the last update time of di. TDS does not
consider the frequency and the size of the documents, which are very important parameters
and influence considerably on the caches performances. Authors in [17] proposed another
cache replacement scheme, called MARS, mobility-aware replacement scheme. It takes
into account various factors that are important when making cache replacement decisions.
MARS uses a cost function for the replacement; it comprises a temporal score and a spatial
score. When the cache of a mobile client becomes full and a new object needs to be cached,
the cost function is used to generate a cost value for each cached object. The object with
the lowest value is evicted from the client’s cache and replaced by the new one. Dimokas
et al. in [18] were also interested on wireless multimedia sensor networks (WMSNs), and
they proposed a cache replacement policy, called NICC associated with a cache consis-
tency mechanism.
2.2 Cooperative Caching
The cooperative caching is based on coordination and sharing of cached data by nodes
belonging to the same area. In this technique, a number of caches work together to cache
more objects collectively, thus, more client’s requests are satisfied without overloading a
single server. Cooperative caching in Ad Hoc networks has been widely studied in
Improving Cache Effectiveness Based on Cooperative Cache...
123
literature, particularly in the context of data accessibility issue [11, 15, 19, 20]. Among the
cache management mechanisms that were proposed within the framework of Ad Hoc
networks, we can quote the work of Du et al. in [11], they proposed a cooperative caching
scheme for on-demand data access applications in MANETs called COOP. This
scheme attempt to discover data sources which induce less communication overhead by
utilizing cooperation zones, historical profiles, and hop-by-hop resolution. In [19], Sailhan
and Issarny proposed a cooperative caching scheme to increase data accessibility by peer-
to-peer (P2P) communication among mobile clients, when they are out of bound of a fixed
infrastructure. Yin et al. in [15], were also interested on data access issue by using
cooperative caching, they proposed three schemes: CacheData which caches the data,
Cachepath which caches the data path and Hybridcache which associate the two schemes
CacheData and Cachepath. In case of Ad Hoc networks with high mobility nodes,
Cachepath or Hybridcache might be unsuitable, because of extra processing overhead
which can be induced. Tang et al. in [20] consider the cache placement problem of
minimizing total data access cost in Ad Hoc networks, they developed efficient strategies
to select data items to cache at each node. The studies achieved by Fiore et al. in [21]
attempt to consider all information of each zone of the MANET as a whole. They proposed
a cooperative scheme called Hamlet integrating cache management strategy that considers
both cases of nodes with large and small cache sizes. Their scheme allows nodes to decide
whether to cache some documents and for how long. Chand et al. in [22] were interested on
caches in MANETs and proposed cooperative scheme called zone cooperative (ZC). As its
name indicates, this scheme is based on zone formed by nodes belonging to the neigh-
borhood of a given client. In [16], authors proposed another cooperative caching strategy in
mobile Ad Hoc networks based on clusters. Chan et al. in [6] studied Energy-saving issues
and formulated the energy-efficient coordinated cache replacement problem (ECORP), it
takes into account a coordinated replacement in the context of a mobile Ad Hoc network.
Group caching (GC) is the technique proposed in [23]. It maintains localized caching status
of 1-hop neighbors. When a data request is received in a mobile host (MH), it forms with
its 1-hop neighbors a group by using the ‘‘Hello’’? message. When caching placement and
replacement need to be performed, the MH selects the appropriate group member to
execute the caching task in the group. In this scheme, the data accessibility is increased but
the energy consumption and constrain of wireless bandwidth are not much considered. In
[24], group-based cooperative caching scheme called (GCC) is proposed. It is based on the
concept of group caching, in which it is allowed for each mobile host and its k-hop
neighbors to form a group. A directory of cached data items is maintained in each mobile
host. Each MH obtains the directories of its k-hop group members through broadcasts in
the group. In this technique, a consequent overhead may be generated by mobile host’s
requests. The study made by Zhang et al. [10] is a continuity of previous works reported in
[25, 26]. Throw these studies, they were interested on the cost efficiency of in-network
caching in Long-Term Evolution by considering two cases, caching in data center only,
and caching in both data center and in-network. They conclude that caching in all network
mobile nodes significantly reduces traffic load and improves latency perceived by end-
users. Chand et al. in [14] proposed cooperative caching scheme, called cluster cooperative
(CC), based on geographic clusters. The Ad Hoc network is partitioned into non-over-
lapping geographic clusters. In this cooperative caching scheme, the clusterization process
is achieved easily because the affiliation of each node to given cluster depends on its
geographic position, however, in the situation of high mobility, an extra processing
overhead can be induced. In the research work of Sureshkumar et al. [27], authors propose
a mobility with clustering based energy efficient cooperative caching method (MCE2C2).
A. Larbi et al.
123
This method integrates clustering algorithm, cache consistency mechanism and cache
replacement policy.
3 Cooperative Cache Management Scheme
In this section, we firstly expose the system model and the presentation of local cache
replacement policy, then we describe the proposed cooperative cache scheme.
3.1 System Model
The system model consists of a set of mobile nodes that can communicate with each other
using Ad Hoc communication protocols. The network topology is thus represented by an
undirected graph G ¼ E;Vð Þ where V is the set of mobile nodes MN1;MN2; . . .;MNn,
Vj j ¼ n and E ¼ V � V . E is the set of links between nodes. Nodes are free to move
randomly, thus the network topology may change rapidly and unpredictably. Each node
acts as a router and is willing to forward data of other nodes. The nodes outside the
communication range of the access points can access the document in the access points
through a number of hops using Ad Hoc network routing protocols. Thus, the document
request is forwarded hop-by-hop until it reaches the access point. During its routing, the
request may flows through many nodes, and it is possible that some of this nodes hold in
their caches a copy of the requested document. In this case, the document is downloaded
from the node that holds a copy of this document, without continuing the routing to access
point.
3.2 Used Local Cache Replacement Policy
The design of the cache replacement policies intended for mobile Ad Hoc networks, or
generally for mobile environments, is strongly influenced by the characteristics of Web
traffic, as well as the specific characteristics of these networks. Among these character-
istics, we can cite document frequencies and their sizes for Web traffic. We can also cite
the distances variation that separates the nodes from/to each other’s due to node’s mobility.
In the following, we present the used cache replacement policy, FSDV Frequency, Size
and Distance based Value. It is based on the calculation of a value for each document
present in the cache. The calculation of this value includes four parameters, (i) access
frequency, (ii) distance in a number of hops between requester node and that one which
answers the request, (iii) time passed since the last update of this same distance and (iv)
document size.
Thus, for each document i, we calculate the value Hi, as expressed in Eq. (1). Then, we
choose document having smallest value for replacement.
Hi ¼fi � di
si � tið1Þ
where
fi : is the access frequency to the document i;
di : is the hops number between requester node for a document i and that one which
answers the request;
si : is the size of the document i;
Improving Cache Effectiveness Based on Cooperative Cache...
123
ti : is the time since the last update of si.
Upon receiving a new document, the cache decision is made locally, for each node of
the network. According to remainder cache free space, if it is insufficient to cache the new
document, cache replacement policy is invoked to liberate space, by selecting one or more
cached documents for eviction. This eviction decision is also made locally.
3.3 Used Cache Consistency Mechanism
Cache consistency mechanisms ensure that each cached document copy is eventually
updated, to reflect changes to the original document. There are several cache consistency
mechanisms currently in use: time-to-live (TTL) fields, client polling and invalidation
protocols. In our system model, we used TTL-based Cache consistency mechanism [28],
however, the CCSP scheme can be associated with any other advanced cache consistency
mechanism. With TTL approach, the origin server assign a TTL-value for each document.
This value is an estimate of the cached document’s lifetime, after this time value, the
document is regarded as invalid, and the cached copy of document is deleted. Thus, the
next request for this document will be satisfied by local cache of another node (through the
CCSP scheme) or, in the worst case, it will be redirected to the origin server.
3.4 Cooperative Cache Scheme Description
The basis of cache’s cooperation is to allow a node in situation of a local cache miss, to be
able to reach this document from other cache nodes.
In our cooperative cache scheme, some nodes proportional to the total number of node
in the network, are elected to play a role of cache services providers. This election is made
according to an algorithm executed periodically to designate another group of SCP, in
order to ensure a good load balancing.
When a given node is elected as SCP, it broadcasts invitation message to the neigh-
boring at one or more hops, in order to request them to send the summary of their caches.
The node receiving multiple invitation messages can reply at most, to three providers.
After sending their respective summaries to SCPs, each node is required to validate its
summary periodically, in all SCPs with which it is associated. If the provider sees that a
given cache summary is still not validated after the end of third validation period, this
summary will be deleted. Additionally, when a node realizes that he is out of communi-
cation range of all the SCPs that have copy of its cache summary, it broadcasts a request to
join another SCP.
The behavior of the cooperative caching scheme based on service cache providers is
shown in Fig. 1. When a node initiates request for document, it first looks for the data item
in its own cache. If there is no copy of requested document in the local cache, the node
sends lookup message to the nearest SCP, this one consults all summaries of local cache
nodes that it holds. If there is an entry for the requested document, so this is a hit, it
responds with the identity of the node having requested document. Otherwise, requester
node contacts another SCP, if still no node that caches the requested document is found,
this document will be recovered directly from the source.
Upon receiving requested document from SCP or from source, requester node checks if
its cache free space is sufficient to cache this document, in which case it is inserted directly
into the cache. Otherwise, the cache replacement policy is invoked to make space for the
new document.
A. Larbi et al.
123
3.5 Election of the Services Cache Providers
As shown in Fig. 2, the election of SCPs is executed periodically following rounds in order
to balance workload, and energy consumption between nodes. Each node calculates Rdiv ið Þas defined in Table 1, those finding zero will be elected as SCPs. After this step, all elected
SCPs send invitation messages to the neighboring at one or more hops, according to j of
SNb i;jð Þ defined in Table 1, in order to request them to send their respective cache summary.
Upon receiving multiple invitation messages, non-SCPs nodes can reply, at most, to three
Requestfor a
document
Consultation of local cache
Checking of freshness
Return the requesteddocument
Send lookup to SCP
Sufficient cache free
space
Replacement
Cache the document
Fresh copy
Yes
Hit
Miss Out-of-date copy
Cache provider Miss
Cache provider Hit
Send lookup to another SCP
Remote Cache provider Hit
Retrieve the document from specific node
Remote Cache provider Miss
Retrieve the document from
the source
Not
Fig. 1 Behavior of the cache receiving a request for a document
Improving Cache Effectiveness Based on Cooperative Cache...
123
providers. After sending their respective summaries to SCPs, non-SCPs nodes are required
to validate their summaries periodically, in all SCPs with which they are associated.
4 Simulation Model
We model our system as a set of mobile nodes (MN) moving in a rectangular area, which
spontaneously formed an Ad Hoc network. Among the mobile nodes, some of them can
directly connect to the Internet; those nodes are considered as Access Points (AP) for the
rest of mobile nodes in the Ad Hoc network. We assume that each access point contains all
requested documents.
Start
Eachcompute
YesNo
Send invitations message to all
nodes in
Wait for invitations message from SCPs
Send its resume at most to three
SCPs
elects itself as SCP
Wait for expiration of round time
Fig. 2 Flowchart of the election mechanism of SCPs
A. Larbi et al.
123
4.1 Notations
According to the system model described above, we can express some notations sum-
marized in Table 1. The evaluation metrics described in Table 1, can be expressed in
Eqs. (2), (3), (4) and (5).
L RHi ¼LH MNij jR MNij j ð2Þ
SCPj RHi ¼SCPjH MNi��
��
R MNij j j ¼ 1. . .3 ð3Þ
L BHi ¼P
dl2LH SDiBdl
P
dk2R SDiBdk
ð4Þ
Table 1 List of simulation notations and their significations
Notations Significations
N Number of nodes in the network
MNi Mobile node i; i ¼ 1. . .N
L Service cache range in number of hops
IDi Node identifier which can take values between 1 and N
pprovider Percentage of cache service providers
Snb i; jð Þ Set neighbors of MNi at j hops
Nround The number of the current round
Rdiv ið Þ The remainder division of IDi þ Nroundð Þ by 1=pprovider
� �
D Number of documents
L RHi Local request hit rate of mobile node i,i ¼ 1. . .N
L BHi Local byte hit rate of mobile node i, i ¼ 1. . .N
SCPj RHi Request hit at the j thSCP of mobile node i, i ¼ 1. . .N, j ¼ 1. . .3
SCPj BHi Byte hit at the j thSCP of mobile node i, i ¼ 1. . .N, j ¼ 1. . .3
G LRH Global request hit rate of Local caches of all mobile nodes
G LBH Global byte hit rate of Local caches of all mobile nodes
G SCPj RH Global request hit at the j thSCP of all mobile nodes, j ¼ 1. . .3
G SCPj BH Global byte hit at the j thSCP of all mobile nodes, j ¼ 1. . .3
Bdi Size of document di (byte)
LH MNi Set of requests that have had a local hit in mobile node i
SCPjH MNi Set of requests that have had a hit at the jth SCP of mobile node i
R MNi Set of requests received by mobile node i
LH SDi Set of all documents concerned by requests that have had a local hit in mobile node i
SCPjH SDi Set of all documents concerned by requests that have had a hit at the jth SCP of mobile nodei
R SDi Set of all documents concerned by requests received by mobile node i
Improving Cache Effectiveness Based on Cooperative Cache...
123
SCPj BHi ¼P
dl2SCPjH SDiBdl
P
dk2R SDiBdk
j ¼ 1. . .3 ð5Þ
Global rates are the average of local rates, so they can be expressed in Eqs. (6), (7), (8)
and (9).
G LRH ¼ 1
N
XN
i¼1
L RHi ð6Þ
G LBH ¼ 1
N
XN
i¼1
L BHi ð7Þ
G SCPj RH ¼ 1
N
XN
i¼1
SCPj RHi j ¼ 1. . .3 ð8Þ
G SCPj BH ¼ 1
N
XN
i¼1
SCPj BHi j ¼ 1. . .3 ð9Þ
4.2 Node Mobility Model
Each mobile node is deployed randomly within the simulation area; we use the Random
Waypoint Movement model to simulate the moving pattern of the mobile nodes. At the
beginning, each mobile node selects a random destination, which is a point on border of
simulation area. To reach its destination, the mobile node moves for a random period of
time and pauses for another random period in alternation, according to the following
sequence: Moving period, pause period and speed. All are distributed uniformly between a
minimum value and a maximum value (Refer to Table 2). After reaching the destination,
the mobile node selects another destination and repeats the movement pattern.
4.3 Performance Evaluation Parameters
In our simulation, we have mainly used three performance parameters. Each of these
parameters is motivated by its direct impact on a specific performance criterion of mobile
Ad Hoc networks. The first one is response time or latency perceived by users (expressed
as latency gain). The latency gain varies whether the request is satisfied from local cache of
requester node or from local cache of another node designated by CCSP scheme. The
second one is request hit rate, which is closely linked to data accessibility performance
criteria. In fact, due to mobility of network nodes, a source of documents could not be
reachable from a given node, so all document requests of this node can be satisfied from a
local cache of this or other network nodes. The third one is byte hit rate. This parameter has
a greater impact on reducing the network workload. Indeed, in situation without using
cache mechanism, upon formulation of request by a given node, the requested document
should transit the network from original server until the requester node.
A. Larbi et al.
123
4.4 Network Workload Trace
In our simulation, the used workload trace is web traffic. This is supported by our study on
mobile Ad Hoc networks (MANETs), indeed the main application used by nodes of
MANETs is the world wide web (WWW). The web traffic has specific characteristics,
much works and studies in scientific literature have focused on workload characterization
[29–32]. These studies are based on statistics and probabilistic methods. As result, several
tools have been developed for the reproduction of web traffic that has characteristics as
close as possible to its original and real characteristics. For our simulation we have used
Table 2 Simulation parameters
Parameters Values
Bandwidth 1 Mbit/s
Number of nodes 100
Inter-request maximum time 3 s
Number of updating request, which can be ignoredby node before deleting its cache summary from SCP
2
Inter-round time period 70 s
Time period of SCPs to request all managed cache nodesto update their summaries
10 s
Inter-invitation time of each SCP to neighbors nodesto send their local cache summaries
5 s
Simulation time 200 s
Cache size 32 Mbits
Simulation area 100 9 100 (m)
Interval of pause time [0–5] (s)
Interval of walking time [2–8] (s)
Interval of walking speed [2–20] (m/s)
Radius of communication range 30 m
The percentage of SCP node (service cache provider) 20%
Synthetic workload parameters
Parameters Values
Parameter of the Zipf law distribution for documentsfrequencies; it is between 0 and 1
0.8
Parameter of the pareto law distribution for documentssize; it is between 0 and 2
1.5
Proportion of distinct documents relative to the total numberof requests; it is between 25 and 40%
30
Proportion of 1-timer documents relative to the total numberof distinct documents; it is between 40 and 70%
50
Smallest document size (bytes) 77
Largest document size (bytes) 1,066,229
Mean document size (bytes) 8840
Improving Cache Effectiveness Based on Cooperative Cache...
123
synthetic web traffic generator tool called ProWGen [33, 34]. The Characteristics of our
synthetic web traffic, generated with this tool are as follows:
1. The web documents frequencies follow Zipf probability law, its detailed parameters
and formula can be found in [33].
2. The web documents sizes follow heavy-tailed probability distribution [13, 23].
Detailed parameters and formula of used probability distribution law can be found in
[33].
3. The proportion of distinct documents, relative to the total number of requests is 30%.
4. The percentage of one-timers documents (documents which are consulted only once)
is 50%.
Table 2 summarizes the important simulation parameters values of our approach.
4.5 System Modeling Description
In our simulation, mobile nodes are the mains entities, around which; we have set up three
processes: (i) mobility process, (ii) cooperation scheme maintenance process and (iii)
document research process. Each process consists of succession of events. Flowchart of
events execution of these three processes is illustrated in Figs. 3, 4 and 5 respectively.
These three processes are executed from beginning to end of the simulation. Their
execution includes one or more mobile nodes. All events created throw this execution are
placed into execution queue or events list. Each event is associated with a timestamp (it is a
precise occurrence time in which the event must be executed).
The execution of each event induces one or more events. The induced events, in turn,
are stamped with their timestamp and are placed in the events list. Simulation loops by
extracting and executing event with the smallest timestamp, from the event list. It runs
until the event list is empty or the simulation stop time is reached.
Frontier of simulation
area is reached
?
Select new random mobility
parametersYes
Move
Moving period expired
No Pause
Pause period expired
Periodic test
Fig. 3 Description of mobility process
A. Larbi et al.
123
5 Simulation Results and Discussion
In this section, we present the simulations experiments of cooperative caching
scheme CCSP by comparing it with SimpleCache scheme (SC) reported in [15] and CC
scheme reported in [14]. In SimpleCache scheme each node has a local cache, and in case
of local cache miss, node’s requests are forwarded to the node source.
In the proposed CCSP scheme, network nodes can be associated to 0, 1, 2 or 3 SCPs at
the same time. The node mobility leads to connections/disconnections of SCPs, which in
Non-SCP node
SCPs election round
Broadcastinvitations
Receive invitations
Receive cache update requests
Receive invitationACK+ cache resume
Send cache update requests
Send cache update
Send invitationACK+ cache resume
Receive cache updates
If cache resume not
updated
After cache update delay
After each election period
After each cache update period
If node is elected as
SCP
No Yes
Send cancel join
Receive cancel join
Yes
If not joined to any SCP
Broadcast joinrequest
Yes
Receive joinrequest
Send joinACK
Receive joinACK
Receive cache resume
Send cache resume
Delete cache resume
SCP node
Fig. 4 Description of cooperation scheme maintenance process
Improving Cache Effectiveness Based on Cooperative Cache...
123
turn varied continually the association number of node with the SCPs. Figure 6 shows the
average rate of these associations. It can be clearly seen that after 89.28% of simulation
time, all nodes are associated to 3 SCPs simultaneously, which allows us to achieve a high
level of cooperation. On the other hand, only 0.42% of simulation time where nodes are not
associated to any SCPs is obtained. In that case, the nodes are not able to cooperate in any
way, with other nodes. So, they can only use their local cache. The global request hit rate
refers to the satisfaction rate of all requests made on the network. In CCSP scheme, a
document request can be satisfied whether from local cache of requester node, or from one
of its three SCPs. Figure 7 shows the part of contribution of each SCP. One can notice
local cache contribution remains insufficient despite the gradually filling up over time of
caches. On the other hand, the contribution of the first SCP (SCP1) is the more important
one in terms of gain in latency. In the beginning, this contribution is widely assisted by the
second SCP (SCP2) and the third SCP (SCP3), but over time, by the progressive filling of
local caches, the contribution of SCP1 becomes more important comparatively to SCP2
and SCP3.
New request
Local hit ?
No Yes
Send a Lookup to its SCP
SCP hit ?
No Yes
Replay with “SCP miss”
YesAll its
SCPs are contacted
?
No
Send request to Access Point
Retrieve document from the cache
Replay with “address of node having requested
document”
Receive a replay
Send a request for the document
Receive the document
Put document in cache
Fig. 5 Description of document research process
A. Larbi et al.
123
To explain in more detail Fig. 8, which represents the gains realized on latency, we
define the following two functions: OSLTime and CCSPTime, describing latency.
OSLTime; CCSPTime : Q ! R
where Q is the set of request and R is the set of real numbers representing latency times.
The first functionOSLTime (Original Source Latency Time) returns for each request the
latency time from the original source of the document. In other words, it returns latency
time of request without using caches. The second one, CCSPTime (Cooperative Caching
based on Service Providers) returns latency time for each request by using CCSP scheme.
The request could be satisfied either from the local cache of requester node or from one of
the three SCPs. Given the foregoing background together, it is now possible to identify the
gains on latency for the local cache, SCP1, SCP2 and SCP3, which are given respectively
by LC TG and SCPj TG where j ¼ 1. . .3, associated formulas are expressed in Eqs. (10)
and (11).
0,42% 2,63%7,66%
89,28%
AvgTime with 0 SCPAvgTime with 1 SCPAvgTime with 2 SCPsAvgTime with 3 SCPs
Fig. 6 Average period of nodesassociated to different number ofSCPs
00,10,20,30,40,50,60,70,80,9
1
5 15 30 50 100 150 200 250 300Req
uest
Hit
Con
tribu
tion
(%)
Simulation Time (s)
Local Request Hit ContributionSCP1 Request Hit ContributionSCP2 Request Hit ContributionSCP3 Request Hit Contribution
Fig. 7 Request hit contributionof local cache, SCP1, SCP2 andSCP3 versus simulation time
Improving Cache Effectiveness Based on Cooperative Cache...
123
LC TG ¼XN
1
X
q2LH MNi
OSLTime qð Þ � CCSPTime qð Þð Þ !
ð10Þ
SCPj TG ¼XN
1
X
q2SCPjH MNi
OSLTime qð Þ � CCSPTime qð Þð Þ
0
@
1
A ð11Þ
where j ¼ 1. . .3:In order to satisfy each document request formulated by a mobile node, local cache is
the first solicited then, at each cache miss, the SCPs cited above, are solicited sequentially.
The SCP can be viewed as a very large cache memory formed by the association of all
local caches of mobile nodes managed by the SCP. It can be seen from Fig. 8, that the most
dominant gain in latency concerns SCP1. So, a very large number of requests that are not
being met by the local cache are redirected to SCP1, where they could be satisfied. The few
missing remaining requests with miss in SCP1 will be redirected to SCP2, and a very few
missing request in SCP2 will be redirected to SCP3. This may explain the higher control of
SCP1TimeGain than SCP2 and SCP3TimeGains. The weak SCP3TimeGain can also be
explained by the fact that document research process may take a long time when the
request is redirected until SCP3. In some cases, this time may be very close to the one
exhausted by the request to get document from the origin server. Therefore, as revealed in
Fig. 8, a low SCP3TimeGain is achieved.
Local request hit rate and local byte hit rate are equivalent to performances, which can
be obtained with network operating without cooperation of local caches. Request hit rate is
closely linked to data accessibility. In fact, due to mobility of network nodes, a source of
documents could not be reachable from a given node, so all document requests of this node
can be satisfied from a local cache of this or other network nodes.
As we can see, in Fig. 9, the gain in request hit rate is very significant by using CCSP
scheme. The request hit rate is more stable over time, it reaches more than 85% in case of
26,34%
56,21%
12,63%
4,82%
Local cache Time gainSCP1 Time gainSCP2 Time gainSCP3 Time gain
Fig. 8 Comparison of timesaving
A. Larbi et al.
123
CCSP scheme and at the same time, that of CC scheme is about 52% and SC request hit
rate was just above the 20%. This represents a gain of about 65 and 35% of satisfied
requests through caches cooperation, compared respectively with CC scheme and SC
scheme. CCSP scheme provides also considerable gain for byte hit rate. A byte hit may be
either local byte hit or remote byte hit, in CCSP scheme a remote hit can occur either in
SCP1, SCP2 or SCP3. In CC scheme, remote hit occurs when requested document is found
within a node belonging to a cluster along the routing path to the source node. In SC byte
hit case, the requested document is recovered without any data being transited through the
network. However, in CCSP byte hit, the requested document should transit from holder
node of that document, to the requester node. This implies that SC byte hit rate will have a
greater impact on reduction of the network load, nevertheless, the remote byte hit rates
impact, mainly the CCSP byte hit rate are not negligible. One can see from Fig. 10 that
when CCSP byte hit rate reached a peak of about 71%, SC byte hit rate is only 14%, and
CC byte hit rate achieves 35%. This makes a gain of 57% by using CCSP scheme against
the use of only SC scheme. These gains were almost realized by the first SCPs (SCP1). The
details of all contributors to these gains are illustrated in Fig. 7. From Figs. 9 and 10, we
can also note that contrary to request hit rate, the byte hit rate recorded slight decrease.
This may be explained by the used cache replacement policy, frequency, size and distance
based value (FSDV) with both CCSP scheme and CC scheme. This policy promotes
0102030405060708090
0 50 100 150 200 250 300 350
Req
uest
Hit
rate
Time in seconds
SC request hit rate CCSP request hit rate CC request hit rateFig. 9 Evolution of request hitrate versus time in seconds
0
10
20
30
40
50
60
70
80
0 50 100 150 200 250 300 350
Byt
e H
it ra
te
Time in seconds
SC byte hit rate CCSP byte hit rate CC byte hit rateFig. 10 Evolution of byte hitrate versus time in seconds
Improving Cache Effectiveness Based on Cooperative Cache...
123
elimination of large documents to free up more space in the cache and therefore increases
the global request hit rate by admission of more documents in the cache.
Figure 11 shows the variation of request hit rate in relation to cache size. We compare
request hit rates provided by CCSP, CC or SC schemes. SC request hit rate is achieved
without any cooperation unlike CCSP scheme and CC scheme. Therefore, Fig. 11 assesses
the gap, between SC request hit rate and those of CCSP and CC schemes. this gap is very
much impacted by the cooperation. It should be noted that the essence of cache cooperation
is related to the limited capacity of local cache memory. Cooperative caching allows
management of all local caches of the same zone by a given node, this provides a global
cache with large storage capacity. Therefore, the number of documents present in global
cache is increased, and request hit rate is increased too. Figure 11 shows a significant
difference between CCSP request hit rate and CC request hit rate, this can be explained by
the fact that unlike CCSP scheme, which use for each node three SCPs at the same time,
the CC scheme use for each node only one CSN. It can be shown also in Fig. 11 that all
request hit rates are constantly evolving with a growth of cache size. It is also interesting to
notice some stability of all request hit rates with a growth of cache size after 96 MB. This
stability may be explained by the satisfaction of a large number of requests from caches
due to the abundant cache free space.
Figures 12 and 13 show the variation of respectively, the number of exchanged mes-
sages in the network and the size of network load, in relation to cache size. The first finding
which can be made in these figures, is that a number of exchanged messages and a size of
network load are closely linked, as illustrated by the curve profiles. The second finding is
the decrease of number of exchanged messages and size of network load by increasing
cache size. Indeed, increasing cache size allows caching more documents, thus, more
requests can be served by intermediate nodes without forwarding lookup messages further
towards the source nodes. Consequently, this will induce significant reduction of network
load. In Fig. 12, it can be observed also that for cache size of 8–16 MB, CC
scheme generates less number of messages than CCSP scheme, but for all others cache size
over than 16 MB, it becomes the other way around. However, we can note that this
difference is not significant, since it varies between 5 and 8% of the total number
exchanged messages.
Figure 14 shows average latencies time of all requests in case of CCSP scheme and CC
scheme, depending on the cache size, from 8 to 128 MB. We note that the curve profiles
are inversely proportional to cache size, which is logical, because, with the growth of cache
0102030405060708090
8 16 32 64 80 96 112 128
Req
uest
Hit
Rat
e (%
)
Cache Size (MB)
SC request hit rate CCSP request hit rate CC request hit rateFig. 11 Evolution of request hitrate versus cache size
A. Larbi et al.
123
size, more and more requests are served from intermediate cache nodes without forwarding
lookup messages further towards the source nodes. Therefore, latency is reduced. We note
in Fig. 14 that, although the difference in terms of request latency time between the two
schemes is clearly visible, but in reality, this difference is not significant, because it does
not exceed 5.5 ms.
Figure 15 shows the variation of time gain depending on the cache size, from 8 to
128 MB. We compare time gain achieved by SC scheme, CCSP scheme and CC scheme.
Despite the slight difference in terms of request latency time between the two schemes
55 000
60 000
65 000
70 000
75 000
80 000
85 000
90 000
0 16 32 48 64 80 96 112 128Num
ber o
f exc
hang
ed m
essa
ges
Cache size (MB)
CCSP message number CC message numberFig. 12 Evolution of number ofexchanged messages versuscache size
100105110115120125130135140145150
0 16 32 48 64 80 96 112 128
Net
wor
k lo
ad s
ize
(MB
)
Cache size (MB)
CCSP network load size CC network load sizeFig. 13 Evolution of networkload size versus cache size
0
4
8
12
16
20
24
0 16 32 48 64 80 96 112 128
Ave
rage
Lat
ensy
Tim
e (M
S)
Cache size (MB)
Average CCSP Latensy Time Average CC Latensy TimeFig. 14 Evolution of averagelatency time versus cache size
Improving Cache Effectiveness Based on Cooperative Cache...
123
CCSP and CC that is illustrated in Fig. 14, in Fig. 15, we note a significant gap between
the time gains of these two schemes. This can be explained by the higher request hit rate
achieved by CCSP scheme, compared with that of CC scheme (This is illustrated in
Fig. 11).
6 Conclusion
The use of cache memories by each node in mobile Ad Hoc networks and its association
with adequate cooperative cache management scheme constitutes one of the best solutions
for improving data accessibility, reducing user’s latency and overall network load. In this
paper, we have proposed the CCSP scheme (cooperative caching scheme based on service
providers).
We were interested on time gain induced by the CCSP scheme in different levels,
namely: local, SCP1, SCP2 and SCP3. The biggest gain was revealed in SCP1. With
increasing in SCP levels, the time gain decreases. The gain remained positive until in
SCP3. Beyond this level, the gain become negative. Indeed, the process of looking for
document in local caches of network nodes may take more time than to bring it directly
from an accessible original server. Simulation results showed that CCSP scheme per-
formed a request hit rate of over than 85%, this performance criteria is closely linked to
data accessibility.
As regards the byte hit rate performance criteria, it was shown that the local byte hit rate
had a greater impact on reduction of the network load. Moreover, the CCSP byte hit rates,
mainly the SCP1 byte hit rate had also non-negligible impact on offloading.
This study showed that the criteria of multi-level document research, adopted by CCSP
scheme is very beneficial for improving data accessibility. Moreover, this proposition
contributes effectively on offloading of mobile Ad Hoc network, in term of volume of
exchanged information. As prospects, it will be interesting to explore possible improve-
ments by using control admission mechanism for each candidate document to be cached. It
will be interesting also to apply coordinated cache replacement policy by considering all
documents cached locally in all nodes of the same neighborhood.
0153045607590
105120135150
8 16 32 48 64 80 96 112 128
Tim
e G
ain
(S)
Cache size (MB)
SC Time Gain CCSP Time Gain CC Time Gain
Fig. 15 Evolution of time gain versus cache size
A. Larbi et al.
123
References
1. Perkins, C. E. (2008). Ad hoc networking (1st ed.). Reading, MA: Addison-Wesley Professional.2. Martikainen, O. E. (2006). Complementarities creating substitutes—Possible paths towards 3g, wlan/
wimax and ad hoc networks. Info, 8(4), 21–32.3. Remy, G., Senouci, S. M., Jan, F., & Gourhant, Y. (2013). Coupling lte and network coding for efficient
dissemination in vehicular networks. Global Information Infrastructure Symposium - GIIS, 2013, 1–7.4. Arlitt, M., Cherkasova, L., Dilley, J., Friedrich, R., & Jin, T. (2000). Evaluating content management
techniques for web proxy caches. SIGMETRICS Performance Evaluation Review, 27(4), 3–11.5. Aggarwal, C., Wolf, J. L., & Yu, P. S. (1999). Caching on the world wide web. IEEE Transactions on
Knowledge and Data Engineering, 11(1), 94–107.6. Chan, E., Li, W., & Chen, D. (2009). Energy saving strategies for cooperative cache replacement in
mobile ad hoc networks. Pervasive and Mobile Computing, 5(1), 77–92.7. Abrams, M., Standridge, C. R., Abdulla, G., Williams, S., & Fox, E. A. (1995). Caching proxies:
Limitations and potentials. Technical report, institution, Blacksburg, VA, USA.8. Reddy, M., & Fletcher, G. P. (1998). Intelligent web caching using document life histories: A com-
parison with existing cache management techniques. In 3rd International WWW caching workshop (pp.35–50).
9. Cao, P., & Irani, S. (1997). Cost-aware www proxy caching algorithms. In Proceedings of the USENIXsymposium on internet technologies and Systems, USITS’97, Berkeley, CA, USA (p. 18). USENIXAssociation.
10. Zhang, N., Kimmerlin, M., Costa-Requena, J., & Hammainen, H. (2016). Cost efficiency of mobile in-network caching. International Journal of Network Management, 26(1), 44–55 (NEM-15-0111.R1).
11. Du, Y., Gupta, S. K. S., & Varsamopoulos, G. (2009). Improving on-demand data access efficiency inMANETs with cooperative caching. Ad Hoc Networks, 7(3), 579–598.
12. Xu, J., Hu, Q., Lee, W.-C., & Lee, D. L. (2004). Performance evaluation of an optimal cachereplacement policy for wireless data dissemination. IEEE Transactions on Knowledge and DataEngineering, 16(1), 125–139.
13. Wu, K.-L., Yu, P. S., & Chen, M.-S. (1996). Energy-efficient caching for wireless mobile computing. InProceedings of the twelfth international conference on data engineering (pp. 336–343).
14. Chand, N., Joshi, R. C., & Misra, M. (2007). Cooperative caching strategy in mobile ad hoc networksbased on clusters. Wireless Personal Communications, 43(1), 41–63.
15. Yin, L., & Cao, G. (2006). Supporting cooperative caching in ad hoc networks. IEEE Transactions onMobile Computing, 5(1), 77–89.
16. Lim, S., Lee, W.-C., Cao, G., & Das, C. R. (2006). A novel caching scheme for improving internet-based mobile ad hoc networks performance. Ad Hoc Networks, 4(2), 225–239.
17. Lai, K. Y., Tari, Z., & Bertok, P. (2004). Location-aware cache replacement for mobile environments.In Global telecommunications conference, 2004. GLOBECOM ’04. IEEE (Vol. 6, pp. 3441–3447).
18. Dimokas, N., Katsaros, D., & Manolopoulos, Y. (2010). Cache consistency in wireless multimediasensor networks. Ad Hoc Networks, 8(2), 214–240.
19. Sailhan, F., & Issarny, V. (2003). Cooperative caching in ad hoc networks. In 4th International con-ference on mobile data management: MDM 2003 (pp. 13–28), Melbourne, Australia.
20. Tang, B., Gupta, H., & Das, S. R. (2008). Benefit-based data caching in ad hoc networks. IEEETransactions on Mobile Computing, 7(3), 289–304.
21. Fiore, M., Casetti, C., & Chiasserini, C. F. (2011). Caching strategies based on information densityestimation in wireless ad hoc networks. IEEE Transactions on Vehicular Technology, 60(5),2194–2208.
22. Chand, N., Joshi, R. C., & Misra, M. (2005). Cooperative cache management in mobile ad hoc net-works. In 2005 2nd Asia Pacific conference on mobile technology, applications and systems (pp. 1–7).
23. Chang, Y.-K., & Ting, Y.-W. (2007). A novel cooperative caching scheme for wireless ad hoc net-works: Groupcaching. 2007 International conference on networking, architecture, and storage (pp.62–68).
24. Ting, I.-W., & Chang, Y.-K. (2013). Improved group-based cooperative caching scheme for mobile adhoc networks. Journal of Parallel and Distributed Computing, 73(5), 595–607.
25. Zhang, N., & Hammainen, H. (2015). Cost efficiency of sdn in lte-based mobile networks: Case finland.In 2015 International conference and workshops on networked systems (NetSys) (pp. 1–5).
26. Naudts, B., Kind, M., Westphal, F. J., Verbrugge, S., Colle, D., & Pickavet, M. (2012). Techno-economic analysis of software defined networking as architecture for the virtualization of a mobilenetwork. In 2012 European workshop on software defined networking (pp. 67–72).
Improving Cache Effectiveness Based on Cooperative Cache...
123
27. Sureshkumar, A., & Samson, R. S. (2016). Swarm and fuzzy based cooperative caching framework tooptimize energy consumption over multimedia wireless sensor networks. Wireless Personal Commu-nications, 90(2), 961–984.
28. Feng, B., Zhou, H., Zhang, H., Jiang, J., & Yu, S. (2016). A popularity-based cache consistencymechanism for information-centric networking. In 2016 IEEE global communications conference(GLOBECOM) (pp. 1–6).
29. Crovella, M. E., & Bestavros, A. (1997). Self-similarity in world wide web traffic: Evidence andpossible causes. IEEE/ACM Transactions on Networking, 5(6), 835–846.
30. Christensen, K. J., & Javagal, N. J. (1997). Prediction of future world wide web traffic characteristics forcapacity planning. International Journal of Network Management, 7(5), 264–276.
31. Arlitt, M., Friedrich, R., & Jin, T. (1999). Workload characterization of a web proxy in a cable modemenvironment. SIGMETRICS Performance Evaluation Review, 27(2), 25–36.
32. Mahanti, A., Williamson, C., & Eager, D. (2000). Traffic analysis of a web proxy caching hierarchy.Network Magazine of Global Internetworking, 14(3), 16–23.
33. Busari, Mudashiru, & Williamson, Carey. (2002). Prowgen: A synthetic workload generation tool forsimulation evaluation of web proxy caches. Computer Networks, 38(6), 779–794.
34. Markatchev, N., & Williamson, C. (2002). Webtraff: A gui for web proxy cache workload modeling andanalysis. In Proceedings on 10th IEEE international symposium on modeling, analysis and simulationof computer and telecommunications systems (pp. 356–363).
Ali Larbi received the engineer degree in Computer Science fromUniversity of Tizi-Ouzou (Algeria) in 2006 and the magister degree inNetworks and Distributed Systems from University of Bejaia (Algeria)in 2009. Currently, he is Ph.D student at University of Bejaia, Memberof LaMOS Research Unit (Modeling and Optimization of Systems).His research interests include mobile networks (Ad Hoc networks,sensors networks…), their architecture, application and resourcemanagement.
Louiza Bouallouche-Medjkoune received the engineer degree inComputer Science from University of Setif (Algeria), the Magisterdegree in Applied Mathematics from University of Bejaıa (Algeria);the Doctorate degree in Computer Science from University of Setif in2006 and the HDR from the University of Constantine in 2009. Sheworks as a teacher at the department of Computer Science ofUniversity of Bejaıa and as a researcher at LaMOS Research Unit ofUniversity of Bejaıa. Bouallouche-Medjkoune is head of the computerscience teaching staff since 2008, and head of the research teamEPSIRT since 2005. She is also head of department of OperationalResearch since January 2010 to July 2015. Her publications haveappeared in various publishers: Taylor&Francis, Springer, AMS, BCS.Her research Interests are: Performance Evaluation of Computer Sys-tems and telecommunication networks, Stability of Systems, MarkovChains, Queueing theory, Computer Networks (ad hoc networks,sensors networks...), Quality of Service.
A. Larbi et al.
123
Djamil Aissani was born in 1956 in Biarritz (Basque Country,France). He started his career at the University of Constantine in 1978.He received his Ph.D in 1983 from Kiev State University (SovietUnion). He is at the University of Bejaia since its opening in1983/1984. Director of Research, Head of the Faculty of Science andEngineering Science (1999–2000), Director of LaMOS (Modeling andOptimization of Systems) Research Unit at Bejaıa University, Scien-tific Head of the Computer Science Doctorate School, he has taught inmany universities (USTHB Algiers, Annaba, Tizi-Ouzou, Rouen,Dijon, ENITA,...). He has published many papers on Markov chains,queueing systems, reliability theory, performance evaluation and theirapplications in such industrial areas as electrical and telecommunica-tion networks and computer systems.
Improving Cache Effectiveness Based on Cooperative Cache...
123